ComfyUI custom nodes for OpenAI's gpt-image-2 (released 2026-04-21) and gpt-image-2.5 Flare and Sunburst (released 2026-09-08), with backwards compatibility for gpt-image-1.5, gpt-image-1, and gpt-image-1-mini.

Nodes

GPT Image 2 — Generate — text-to-image
GPT Image 2 — Edit — image-to-image with up to 9 reference images + optional mask
GPT Image 2 — Estimate Cost — dry-run token + USD estimator (no API call)

All three live under the OpenAI/GPT-Image-2 category in the right-click → Add Node menu.

Install

Option A — ComfyUI Manager

Manager → Install via Git URL:

https://github.com/graedance/comfyui-gpt-image-2

Option B — git clone

cd ComfyUI/custom_nodes
git clone https://github.com/graedance/comfyui-gpt-image-2.git
cd comfyui-gpt-image-2
pip install -r requirements.txt

For the portable Windows ComfyUI build, replace pip with:

..\..\python_embeded\python.exe -m pip install -r requirements.txt

Restart ComfyUI.

API key

Set once as an environment variable (recommended) so you never paste it into a workflow:

# macOS / Linux
echo 'export OPENAI_API_KEY="sk-..."' >> ~/.zshrc
source ~/.zshrc

# Windows
setx OPENAI_API_KEY "sk-..."

Or paste it into the optional api_key input on any node — useful for one-off runs but it'll be saved into the workflow JSON.

Models

| Model ID | Notes | |---|---| | gpt-image-2.5-flare | Default. OpenAI's recommended 2.5 model: higher quality than gpt-image-2 at about half the latency. Released 2026-09-08. | | gpt-image-2.5-sunburst | Premium 2.5 model: slower, tighter control across edits. | | gpt-image-2.5-flare-2026-09-08 | Pinned Flare snapshot. | | gpt-image-2.5-sunburst-2026-09-08 | Pinned Sunburst snapshot. | | gpt-image-2 | Previous generation. | | gpt-image-2-2026-04-21 | Pinned snapshot for reproducibility. | | gpt-image-1.5 | Legacy. | | gpt-image-1 | Legacy. | | gpt-image-1-mini | Cost-optimized legacy. |

Sizes

gpt-image-2 accepts any resolution within these constraints: max edge ≤ 3840 px, both edges multiples of 16, long:short ≤ 3:1, total pixels 655,360–8,294,400. Common values exposed in the dropdown:

auto, 1024x1024, 1024x1536, 1536x1024, 1792x1024, 1024x1792, 2048x1152, 2048x2048, 3840x2160, 2160x3840

gpt-image-1.x is restricted to 1024x1024, 1024x1536, 1536x1024, auto. Picking a 2K/4K size on a 1.x model will cause the API to reject the request.

Quality

auto, low, medium, high, plus xhigh and max on the 2.5 models only. Older models reject xhigh and max.

Cost estimator

The Estimate Cost node returns five outputs:

text_tokens (INT)
image_in_tokens (INT)
image_out_tokens (INT)
usd_cost (FLOAT)
summary (STRING) — wire to a ShowText node for a human-readable breakdown

Example output for 4× high 1024² gpt-image-2:

gpt-image-2 | 4× 1024x1024 high
  text in:       18 tok  → $0.0001
  image in:       0 tok  → $0.0000
  image out:  28132 tok  → $0.8440
  total:                  $0.8441

Pricing is hard-coded from OpenAI's published rates (text in $5/M, image in $8/M, cached $2/M, image out $30/M for gpt-image-2). The 2.5 models bill at the same token rates. OpenAI has not published output-token counts for xhigh or max, so the estimator prices them as high and labels the figure as a floor. Toggle cached_input to apply the cached image rate.

License

MIT.